ART
Agent Reinforcement Trainer: train multi-step agents for real-world tasks using GRPO. Give your agents on-the-job training. Reinforcement learning for Qwen3.6, GPT-OSS, Llama, and more!
파일 탐색기
최종 버전 다운로드 (.zip)- SKILL.md
- SKILL.md
- SKILL.md
- skills
- build-gpu-image.yml
- create-draft-release.yml
- package-install.yml
- prek.yml
- release.yml
- trainer-rank-gpu.yml
- CODEOWNERS
- accuracy-comparison.svg
- accuracy-training-progress.svg
- win_rate_over_time.png
- accuracy-comparison.svg
- accuracy-training-progress.svg
- accuracy-comparison.svg
- accuracy-training-progress.svg
- hallucinated-facts-comparison.svg
- has-conservative-bias-comparison.svg
- has-liberal-bias-comparison.svg
- includes-all-facts-comparison.svg
- accuracy-comparison.svg
- accuracy-training-progress.svg
- accuracy-comparison.svg
- accuracy-training-progress.svg
- accuracy-comparison.svg
- accuracy-training-progress.svg
- ART_E_graphs.png
- ART_E_pill.png
- ART_header.png
- ART_logo.png
- ART_pill.png
- Colab_pill.png
- Discord.png
- Documentation_pill.png
- Header_separator.png
- MCP_RL_diagram.svg
- Train_pill.png
- math-vista.ipynb
- math-vista.py
- benchmark_inference.py
- gemma3.py
- prompts.json
- qwen3_try.ipynb
- qwen3_try.py
- dataset.jsonl
- distillation.py
- sft-from-file.py
- sft-warmup.py
- generate_images.py
- train.ipynb
- bench_cute_grouped_lora.py
- demo_logging.py
- megatron_review_perf.py
- profile.ipynb
- run_qwen3_5_localbackend_yes_no_maybe.py
- run_qwen3_5_megatron_yes_no_maybe.py
- run_yes_no_maybe_kl_advantage.py
- run_yes_no_maybe_kl_advantage_tinker.py
- tau-bench-minimal.py
- trainer_rank.py
- trainer_rank_check.py
- trainer_rank_checkpoint_acceptance.py
- trainer_rank_diag.py
- trainer_rank_fast_check.py
- trainer_rank_support.py
- tune_quack_lora_tiles.py
- yes-no-maybe-fork-pipeline.py
- yes-no-maybe-fork-pipeline.sky.yaml
- yes-no-maybe-fork.py
- yes-no-maybe-fork.sky.yaml
- yes-no-maybe-kl-advantage-tinker.py
- yes-no-maybe-kl-advantage.py
- yes-no-maybe-megatron.py
- yes-no-maybe-metrics.py
- yes-no-maybe.ipynb
- yes-no-maybe.py
- yes_no_maybe_trainability.py
- art-gpu.Dockerfile
- gspo.mdx
- additional-histories.mdx
- checkpoint-deletion.mdx
- checkpoint-forking.mdx
- mcp-rl.mdx
- tracking-metrics.mdx
- art-backend.mdx
- art-client.mdx
- ruler.mdx
- sft-training.mdx
- training-loop.mdx
- about.mdx
- faq.mdx
- installation-setup.mdx
- notebooks.mdx
- quick-start.mdx
- art-loop.webp
- dark.svg
- light.svg
- favicon.webp
- forked-run.webp
- open-deep-research-progress.png
- ruler-results.png
- langgraph-integration.mdx
- openenv-integration.mdx
- backend-first-training-api.md
- glossary.mdx
- models.mdx
- open-deep-research.mdx
- summarizer.mdx
- .gitignore
- analytics.js
- docs.json
- package.json
- pnpm-lock.yaml
- README.md
- style.css
- display_benchmarks.ipynb
- generate_benchmarks.py
- rollout.py
- train.py
- utils.py
- codenames_words.json
- dictionary.json
- prompts.txt
- puzzles.json
- reference_grpo_trainer.py
- train.py
- utils.py
- __init__.py
- checks.py
- display_benchmarks.ipynb
- experiments.py
- find_articles.py
- generate_benchmarks.py
- rollout.py
- scenarios.py
- train.py
- utils.py
- .gitignore
- main.py
- pyproject.toml
- README.md
- test_scraper.py
- uv.lock
- mcp_alphavantage.ipynb
- mcp_balldontlie.ipynb
- generate_benchmarks.py
- __init__.py
- checks.py
- mcp_server.py
- rollout.py
- scenario_generator.py
- train.py
- utils.py
- train.jsonl
- val.jsonl
- __init__.py
- README.md
- scenarios.jsonl
- server.py
- server_params.py
- train.jsonl
- val.jsonl
- __init__.py
- README.md
- server.py
- server_params.py
- __init__.py
- pyproject.toml
- README.md
- server.py
- server_params.py
- all_experiments.py
- pyproject.toml
- README.md
- test_scenario_generation.py
- uv.lock
- .gitignore
- temporal-clue-7b-async.ipynb
- temporal-clue-7b.ipynb
- temporal-clue.py
- display-benchmarks.ipynb
- game_utils.py
- rollout.py
- tic-tac-toe.py
- deploy_step.py
- game_utils.py
- gather_trajectory_groups_by_index.py
- rollout.py
- train.py
- train_o4_mini.py
- benchmarking_comparison_models.py
- openenv_echo.py
- prisoners-dilemma.ipynb
- rock-paper-tool-use.ipynb
- GPL-3.0.txt
- LGPL-3.0.txt
- backend.vcs.txt
- apply_ci_uv_build_overrides.py
- build_and_push_uv_cache.sh
- compute_uv_fingerprint.py
- github_release_assets.py
- trainer-rank-checkpoint-acceptance.sky.yaml
- trainer-rank-gpu-tests.sh
- trainer-rank-gpu.sky.yaml
- build-gpu-image.sh
- build_package.py
- bump_version.py
- deploy-model.py
- kill-gpu-processes.sh
- migrate-s3-checkpoints.py
- publish.sh
- setup.sh
- __init__.py
- engine.py
- get_model_config.py
- model.py
- openai_server.py
- sequence_lengths.py
- train.py
- validate.py
- __init__.py
- llm_wrapper.py
- logging.py
- message_utils.py
- __init__.py
- adapter_leases.py
- backend.py
- checkpoints.py
- service.py
- __init__.py
- default_tools.py
- generate_scenarios.py
- types.py
- allocator.cu
- allocator.cuh
- hybrid_ep_backend.cuh
- ibvcore.h
- NCCL_LICENSE.txt
- topo_detection.cuh
- internode.cuh
- internode_doca.cu
- internode_doca.cuh
- internode_nixl.cu
- internode_nixl.cuh
- intranode.cu
- intranode.cuh
- nixl_connector.cu
- nixl_connector.h
- executor.cu
- executor.cuh
- allgather.cu
- allgather.cuh
- permute.cu
- permute.cuh
- compiler.cu
- compiler.cuh
- config.cuh
- coordinator.cuh
- hybrid_ep.cu
- hybrid_ep.cuh
- pybind_hybrid_ep.cu
- utils.cuh
- __init__.py
- hybrid_ep_buffer.py
- hybrid_ep_toolchain.py
- LICENSE
- pyproject.toml
- setup.py
- VERSION
- __init__.py
- block_mask.py
- builder.py
- comm.py
- core_attention.py
- executor.py
- layout_index.py
- loss_inputs.py
- range_ops.py
- runtime.py
- types.py
- __init__.py
- precision_aligned_ops.py
- tilelang_import.py
- tilelang_indexer.py
- tilelang_indexer_bwd.py
- tilelang_indexer_fwd.py
- tilelang_sparse_mla.py
- tilelang_sparse_mla_bwd.py
- tilelang_sparse_mla_fwd.py
- __init__.py
- bridge.py
- compressor.py
- deepseek_v4.py
- dequant.py
- encoding.py
- hf_config.py
- hyper_connection.py
- layer.py
- lora.py
- rope.py
- spec.py
- tokenizer.py
- utils.py
- v4_indexer.py
- __init__.py
- attention.py
- compiled.py
- flash_dlse_patch.py
- __init__.py
- conv_gelu.py
- fla_cp.py
- fla_cp_kernels.py
- gdn_prefix_tree.py
- l2norm.py
- layout.py
- operator.py
- segment_layout.py
- __init__.py
- cute_grouped_lora_quack.py
- __init__.py
- default_dense.py
- dsv4.py
- gemma4.py
- gpt_oss.py
- llama3.py
- qwen3_5.py
- qwen3_common.py
- qwen3_dense.py
- qwen3_moe.py
- __init__.py
- discovery.py
- internal_padding.py
- lora_disk.py
- registry.py
- spec.py
- tokenizer.py
- __init__.py
- bridge_runtime.py
- client.py
- jobs.py
- runtime_env.py
- te_cutlass_grouped_gemm.py
- __init__.py
- compile.py
- finalize_grads.py
- microbatches.py
- model_chunks.py
- offload.py
- sft_batches.py
- streaming_weight_offload.py
- trace.py
- weight_offload.py
- __init__.py
- adapter_export.py
- lora_publish.py
- merged_weight_export.py
- param_name_canonicalization.py
- __init__.py
- backend.py
- compile_workarounds.py
- hybrid_ep_setup.py
- lora.py
- megatron_patches.py
- migrations.py
- optimizer_state.py
- prefix_tree.py
- prefix_tree_packing.py
- prefix_tree_state.py
- provider.py
- routing_replay.py
- runtime_config.py
- service.py
- setup.sh
- train.py
- __init__.py
- binary_prefix_tool_pipeline.py
- checkpoint_retention.py
- state.py
- status.py
- trainer.py
- types.py
- yes_no_maybe_pipeline.py
- __init__.py
- attachment.py
- autotune.py
- config.py
- store.py
- worker_controller.py
- __init__.py
- inputs.py
- moe_routing.py
- pack.py
- policy_spans.py
- response_masking.py
- tokenize.py
- vllm_tokens.py
- __init__.py
- ruler.py
- __init__.py
- backend.py
- client.py
- __init__.py
- client.py
- rollout.py
- test_kl_advantage.py
- test_step_skipping.py
- __init__.py
- backend.py
- client.py
- prefix_cache.py
- renderers.py
- server.py
- service.py
- __init__.py
- backend.py
- data.py
- __init__.py
- _checkpoint.py
- _impl.py
- _lora_export.py
- _telemetry.py
- topk.py
- __init__.py
- aiohttp.py
- core.py
- httpx.py
- requests.py
- __init__.py
- _compact.py
- _compat.py
- _history.py
- _protocols.py
- _scope.py
- _selection.py
- _serialization.py
- _tokenize.py
- tensors.py
- __init__.py
- patches.py
- __init__.py
- service.py
- train.py
- __init__.py
- percentage_comparison_bar_chart.py
- training_progress_chart.py
- aggregate_trajectories.py
- filter_model_split.py
- load_trajectories.py
- log_constant_metrics_wandb.py
- pull_model_trajectories.py
- types.py
- __init__.py
- common.py
- legacy.py
- together.py
- wandb.py
- calculate_step_metrics.py
- display_image_grid.py
- generate_comparison_table.py
- generate_line_graphs.py
- load_benchmarked_models.py
- types.py
- __init__.py
- benchmark_rollout.py
- chat_template.py
- convert_moe_lora.py
- deploy_model.py
- format_message.py
- get_model_step.py
- get_repo_root_path.py
- group_aggregate.py
- iterate_dataset.py
- lifecycle.py
- limit_concurrency.py
- litellm.py
- log_http_errors.py
- logging.py
- managed_process.py
- model_config.py
- network.py
- output_dirs.py
- record_provenance.py
- retry.py
- s3.py
- s3_checkpoint_utils.py
- sft.py
- strip_logprobs.py
- suppress_litellm_serialization_warnings.py
- trajectory_logging.py
- trajectory_migration.py
- wandb_sdk.py
- __init__.py
- nccl.py
- packed_tensor.py
- __init__.py
- _backend_training.py
- adapter_leases.py
- api_costs.py
- backend.py
- batches.py
- cli.py
- costs.py
- errors.py
- gather.py
- guided_completion.py
- loss.py
- metrics.py
- metrics_taxonomy.py
- model.py
- openai.py
- py.typed
- serving_capabilities.py
- types.py
- vllm_route_transport.py
- vllm_runtime.py
- yield_trajectory.py
- __init__.py
- move.py
- traceback.py
- .gitignore
- __init__.py
- megatron_attention_oracle_harness.py
- megatron_attention_oracle_worker.py
- test_attention_packed_vs_flattened.py
- test_megatron_attention_oracle_correctness.py
- .gitignore
- __init__.py
- artifacts.py
- cases.py
- distributed_grad.py
- distributed_init.py
- layout_reference.py
- metrics.py
- oracles.py
- packed_layout.py
- parser_import.py
- real_gdn_oracle.py
- test_fla_cp_native_recurrent.py
- test_gdn_conv_gelu.py
- test_gdn_cp_layout_distributed.py
- test_gdn_cp_packed_correctness.py
- test_gdn_cp_train_prepare.py
- test_gdn_planner_runtime_model.py
- test_qwen35_full_model_cp1_packed_vs_flattened.py
- test_qwen35_gdn_topology_oracle.py
- test_real_gdn_cp1_packed_vs_flattened.py
- test_real_gdn_native_fla_cp.py
- test_real_gdn_tp_lora.py
- __init__.py
- merged_vllm_serving.py
- native_vllm_lora.py
- test_dynamic_lora_slots.py
- test_lora_disk_codecs.py
- test_lora_module_loading.py
- test_merged_weight_export.py
- test_weight_transfer_bootstrap_contract.py
- __init__.py
- chat_template_rollout.py
- forward_trace.py
- fp32_grouped_gemm.py
- gdn_fp32_reference.py
- gdn_trace_uids.py
- hf_parity.py
- hf_parity_canonicalization.py
- hf_parity_worker.py
- lora_coverage.py
- oracle_harness.py
- oracle_worker.py
- packing_invariance.py
- prefix_tree_workloads.py
- routing_replay_bundle.py
- test_compile_flags.py
- test_dsv4_real_path_correctness.py
- test_hf_parity.py
- test_hf_parity_invariants.py
- test_inputs.py
- test_internal_padding.py
- test_lora_oracle_correctness.py
- test_oracle_harness_invariants.py
- test_packing_invariance.py
- test_provider_support.py
- test_workflow.py
- trace_uids.py
- validation_spec.py
- workflow.py
- workflow_resources.py
- workflow_stage_worker.py
- __init__.py
- bundle.py
- trace.py
- .gitignore
- __init__.py
- artifacts.py
- conftest.py
- README.md
- test_art_import_boundary.py
- test_art_separation_contract.py
- test_client.py
- test_live_local_backend_smoke.py
- test_live_megatron_backend_smoke.py
- test_live_runtime_server_smoke.py
- test_runtime_launcher.py
- test_runtime_project_isolation.py
- test_service_runtime_boundary.py
- .gitignore
- sitecustomize.py
- __init__.py
- artifacts.py
- conftest.py
- output_parity.py
- real_path.py
- test_config.py
- test_live_real_path_output_parity.py
- test_output_parity_invariants.py
- workflow_stage.py
- __init__.py
- test_config.py
- test_live_length_trainability.py
- test_live_yes_no_trainability.py
- yes_no_trainability.py
- __init__.py
- test_streaming_offload_oracle.py
- test_streaming_offload_trainability.py
- __init__.py
- artifacts.py
- conftest.py
- metrics.py
- test_optimizer_state_contract.py
- __init__.py
- test_live_api_cost.py
- test_lora_quack_cutover.py
- test_multi_checkpoint_training.py
- test_pipeline_lifecycle_contract.py
- test_pipeline_localbackend_dedicated.py
- test_provenance.py
- test_push_and_fork.py
- test_qwen_tool_parser_contract.py
- test_sft_last_assistant_tokenization.py
- test_tinker_native_backend.py
- test_vllm_contract.py
- __init__.py
- chat_template_conformance_cases.py
- qwen_tool_parser_contract.json
- test_capture.py
- test_compact_serialization.py
- test_history.py
- test_tensorized_models.py
- test_tokenize.py
- test_tokenized_models.py
- test_auto_trajectory.py
- test_benchmarking_loader.py
- test_checkpoint_retention.py
- test_dedicated_config.py
- test_dsv4_vllm_runtime_patches.py
- test_exchange_training_model_selection.py
- test_frontend_logging.py
- test_langgraph_llm_wrapper.py
- test_langgraph_logging.py
- test_local_sft.py
- test_megatron_reference_logprobs.py
- test_merged_weight_names.py
- test_metric_routing.py
- test_metrics_builder.py
- test_metrics_taxonomy.py
- test_model_config.py
- test_model_openai_client_costs.py
- test_model_support_import_boundary.py
- test_moe_routing_real_path.py
- test_moe_routing_replay.py
- test_mp_actors.py
- test_multi_checkpoint_inference.py
- test_openai_stream.py
- test_pipeline_autotuner.py
- test_pipeline_trainer_batching.py
- test_pipeline_trainer_local_backend.py
- test_pipeline_trainer_metrics.py
- test_prefix_cache.py
- test_prefix_tree.py
- test_prefix_tree_attention_builder.py
- test_prefix_tree_grad_parity.py
- test_prefix_tree_packing.py
- test_preprocessing_tokenize.py
- test_qwen_tool_parser_config.py
- test_ruler_metrics.py
- test_serverless_adapter_lease.py
- test_serverless_pipeline_trainer_compat.py
- test_sft.py
- test_strip_logprobs.py
- test_tau_bench_client.py
- test_tinker_native_exchanges.py
- test_tinker_native_kl.py
- test_tinker_renderers.py
- test_tokenize_trajectory_groups.ipynb
- test_track_api_cost.py
- test_train_inf_mismatch_logical_map.py
- test_trainer_rank_check.py
- test_trainer_rank_custom_tensors.py
- test_trainer_rank_telemetry.py
- test_trainer_rank_validation.py
- test_trainer_rank_weird_shapes.py
- test_trajectory_copy.py
- test_trajectory_parquet.py
- test_unsloth_autocast_dtype.py
- test_unsloth_metrics.py
- test_vllm_lora_delta.py
- test_yield_trajectory.py
- __init__.py
- integration.py
- test_backend_train_api.py
- __init__.py
- binary_routes.py
- dedicated_server.py
- dsv4_patches.py
- gemma4_moe_lora_patch.py
- lora_delta.py
- metrics.py
- patches.py
- policy_spans.py
- pyproject.toml
- uv.lock
- .dockerignore
- .env.example
- .gitattributes
- .gitignore
- .pre-commit-config.yaml
- .python-version
- .skyignore
- AGENT.md
- CLAUDE.md
- CONTRIBUTING.md
- LICENSE
- pyproject.toml
- README.md
- skypilot-config.yaml
- THIRD-PARTY-NOTICES
- uv.lock
// repository documentation
Was this content helpful?
(0 ratings)
